计算机自适应测验分阶段应用能力估计方法的模拟分析  

Comparison between Ability Estimation Methods under Computerized Adaptive Testing

作　　者：简小珠 Jian Xiaozhu(Department of Education,Guangxi Normal University,School of Education,Jinggangshan University,Ji'an,Jiangxi,343000)

机构地区：[1]广西师范大学教育学部,江西吉安343000 [2]井冈山大学教育学院,江西吉安343000

出　　处：《考试研究》2023年第2期56-67,共12页Examinations Research

基　　金：国家社会科学基金后期资助项目(编号:21FJKB021);江西省社会科学研究规划项目(计算机化自适应测验技术发展分析与实测应用,编号19JY02)的研究成果。

摘　　要：概述计算机自适应测验的极大似然估计方法、极大后验估计方法、期望后验估计方法及其变式以及优缺点。在CAT测试初始、中间、最终阶段分别设计不同的能力估计方法并进行模拟研究。结果显示,CAT的初始、中间、最终阶段同时使用MLE或Biweight、EAPE-U(-4,4)方法,各个能力水平的被试均能被准确测量;CAT的初始、中间、最终阶段中使用EAPE-N(0,1)方法或EAPE-N(0,2)方法,则高能力被试出现一定程度低估现象,低能力被试出现一定程度高估现象,而且所有被试的能力估计值呈现向能力量尺的中间靠拢的趋势。This paper summarizes three main types of computerized adaptive testing(CAT)ability estimation methods,namely,maximum likelihood estimation method(MLE),maximum posterior estimation method(EAPE),expectation posterior estimation method(MAPE)and their variants,and discusses their advantages,disadvantages and applicable situations. In this paper,through CAT simulation design,different ability estimation methods are used in the CAT testing process and the final stage of CAT respectively,and the measurement attributes of the ability estimation methods in the CAT testing process are analyzed. It is found that under the methods of MLE,Biweight and EAPE-U(-4,4),CAT could achieve accurate measurement for all ability levels of the subjects. In the process of CAT test,when the ability estimation method of subjects is EAPE-N(0,1)or EAPE-N(0,2),and the final ability estimation method is EAPE-N(0,1)or EAPE-N(0,2),high-ability subjects will underestimate to a certain extent,the low-ability subjects overestimated to a certain extent,and the ability estimation was close to the middle. In addition,as long as EAPE-N(0,1)or EAPE-N(0,2)method is used in part of CAT stage,and other ability estimation methods such as MLE are used in other stages,the RMSE of intermediate ability subjects will be relatively small,while the RMSE of high-ability and low-ability subjects will be relatively large.

关 键 词：CAT 极大似然估计 极大后验估计 期望后验估计 

分 类 号：G424.74[文化科学—课程与教学论]